River: A Tightly-coupled Radar-inertial Velocity Estimator

0. Preliminaries

If you use River in a scientific publication, please cite the following paper 👇:

• S. Chen, X. Li*, S. Li, Y. Zhou and S. Wang, "River: A Tightly-Coupled Radar-Inertial Velocity Estimator Based on Continuous-Time Optimization," in IEEE Robotics and Automation Letters (RA-L), 2024. [paper] [code] [video]

News: Calibrate Your Radar-Inertial Sensor Suite »

• supports one-shot multi-radar multi-IMU spatial and temporal determination;
• not require any additional artificial infrastructure or prior knowledge;
• capable of accurate and consistent calibration;

Todo List »

• double free or corruption (out) Issue #3
• handle standstill motion (stationary) case (directly quit estimator, currently).
• support multi-radar multi-inertial velocity estimation.
• support online radar-inertial calibration.

1. Overview

Continuous and reliable ego-velocity information is significant for high-performance motion control and planning in a variety of robotic tasks. While linear velocities as first-order kinematics can be simultaneously estimated with other states or explicitly obtained by differentiation from positions in ego-motion estimators such as odometers, the high coupling leads to instability and even failures when estimators degenerate. To this end, we present River: an accurate and continuous linear velocity estimator that efficiently fuses high-frequency inertial and radar target measurements based on continuous-time optimization.

Our accompanying videos are now available on YouTube (click below images to open) and Bilibili.

---

---

2. Build River

2.1 Preparation

• install ROS1 (Ubuntu 20.04 is suggested, Ubuntu 18.04 (ros melodic) is also available):

  sudo apt install ros-noetic-desktop-full
  echo "source /opt/ros/noetic/setup.bash" >> ~/.bashrc
  source ~/.bashrc

  Requirements: ROS1 & C++17 Support

• install Ceres:

  see the GitHub Profile of Ceres library, clone it, compile it, and install it. Make sure that the version of Ceres contains the Manifold module. (Ceres version equal to 2.2.0 or higher than that)

• install Sophus:

  see the GitHub Profile of Sophus library, clone it, compile it, and install it. Set optionSOPHUS_USE_BASIC_LOGGING ON when compile (cmake) the Sophus library, this would avoid to involve fmt logger dependency (as the following spdlog would use internal fmt too, which may lead to conflict).

• install magic-enum:

  see the GitHub Profile of magic-enum library, clone it, compile it, and install it.

• install Pangolin:

  see the GitHub Profile of Pangolin library, clone it, compile it, and install it.

• install spdlog:

  see the GitHub Profile of spdlog library, clone it, compile it, and install it. Though this library can be installed by sudo apt-get install libspdlog-dev, the version is too old to support river, thus installing from the source is recommended.

• install cereal, yaml-cpp:

  sudo apt-get install libcereal-dev
  sudo apt-get install libyaml-cpp-dev

2.2 Clone and Compile River

• create a ros workspace if needed and clone River to src directory as river:

  mkdir -p ~/River/src
  cd ~/River/src
  
  git clone --recursive https://github.com/Unsigned-Long/River.git river

  change directory to 'river', and run 'build_thirdparty.sh'.

  cd river
  chmod +x build_thirdparty.sh
  ./build_thirdparty.sh

  this would build 'tiny-viewer' and 'ctraj' libraries.

• Prepare for thirdparty ros packages:

  clone ros packages 'ainstein_radar', 'ti_mmwave_rospkg', 'serial', 'sbg_ros_driver' to 'river/..' (directory at the same level as river):

  cd ..
  git clone https://github.com/AinsteinAI/ainstein_radar.git
  git clone https://github.com/Unsigned-Long/ti_mmwave_rospkg.git
  git clone https://github.com/wjwwood/serial.git
  git clone https://github.com/SBG-Systems/sbg_ros_driver.git

  then change directory to the ros workspace to build these packages:

  cd ..
  catkin_make -DCATKIN_WHITELIST_PACKAGES="ainstein_radar;ti_mmwave_rospkg;serial;sbg_driver"

  Note that these packages will depend on many other ros packages, you need to install them patiently.

• compile River:

  # generate the ros self-defined message: 'RiverState'
  catkin_make river_generate_messages
  # compile river package
  catkin_make -DCATKIN_WHITELIST_PACKAGES=""

3. Launch River

Attention: to create a virtual reality (VR) perspective of the IMU (left window view in runtime), you have to change the model file path in configure field (Preference::ObjFileForDisplay) to {root path}/river/model/river.obj. For a better VR perspective, you can design your own simulation scenario using Blender and export it as an obj file, then pass it to the configure file.

3.1 Simulation Test

datasets, launch, result visualization

3.2 Real-world Experiments

datasets, launch, result visualization

3.3 Skip Tutorial

Find a configure file in river/dataset, then change the fields in the configure files to be compatible with your dataset (there are detailed comments for each field). You only need to change a few fields related to io (input and output), perhaps some additional fields related to optimization.

Then give the path of your configuration file to the launch file of river in folder river/launch (handheld, xrio, or simu-test folder), Then, we launch 'river':

roslaunch river {the-launch-filename}.launch

The corresponding results would be output to the directory you set in the configure file when solving finished (or interrupted).